AnttiHero

AnttiHero / lavern

Public

An agentic law firm. Yours. 67 specialist AI agents that review documents through evidence-backed debate, with mandatory human gates and a 10-pass verification loop. Apache 2.0.

lavern.ai agentic-ai ai-agents anthropic claude contract-review
37
14
85% credibility
Found May 20, 2026 at 38 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
TypeScript
AI Summary

Lavern is an open-source multi-agent legal analysis system that simulates a team of 67 specialized AI lawyers working together to review contracts, agreements, and legal documents. The system uses a debate protocol where different specialists analyze documents, cite specific text evidence, challenge each other's findings, and run multiple verification layers before presenting conclusions. It includes human approval checkpoints for critical decisions and produces complete audit trails. Users can run it locally (free, private), through Anthropic's Claude AI (paid, cloud), or through Mistral AI (EU-based, GDPR-friendly). The project includes a React dashboard, automated document monitoring capabilities, and a macOS menu bar app. It comes with 1,677 tests and is licensed under Apache 2.0. The creators explicitly note this is a 'collection of ideas' and 'source of inspiration' rather than a finished product, and acknowledge that the quality of legal output has not been independently validated against real legal work.

How It Works

1
📄 You upload a legal document

You drop in a contract, agreement, or legal document you need reviewed.

2
💬 You describe what you need

In plain language, you explain your question or concern about the document.

3
⚡ A team of specialists springs into action

Multiple AI specialists analyze your document simultaneously, citing exact text from your document.

4
🔍 They debate and challenge each other

The specialists check each other's work, challenge weak findings, and verify evidence.

5
Critical findings need your approval
✅
You approve the finding

The work continues with your blessing.

✏️
You override or redirect

You provide guidance and the team adjusts course.

6
📋 You receive a complete deliverable

A polished document review with findings, citations, confidence scores, and an audit trail.

🎉 You have professional-grade legal analysis

A thorough review that would typically require a team of specialists, now delivered on your machine.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 38 to 37 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is lavern?

Lavern is an agentic legal system built in TypeScript that simulates a multi-specialist law firm using 67 AI agents working together through a structured debate protocol. Instead of relying on a single LLM to handle legal documents, Lavern routes work through specialized agents—contract reviewers, tax counsel, IP specialists, litigation partners—that must cite specific text from parsed documents when making findings. The system runs three layers of verification (evaluator gate, adversarial debate, 10-pass pipeline) and pauses at human gates before delivering critical conclusions. You can run it against Anthropic Claude, Mistral AI for EU compliance, or locally via Ollama. The React dashboard shows agents debating in real time, and Clawern mode processes documents autonomously on a 30-minute heartbeat with Telegram/email alerts.

Why is it gaining traction?

The multi-agent debate architecture is the hook. Rather than asking a single model to be a "junior associate," Lavern forces agents to challenge each other with cited evidence, creating an audit trail that shows *why* a finding was made. The 10-pass verification loop and precedent board (which accumulates cross-document patterns) address a real pain point: legal work product quality is hard to measure, and this system at least structures the verification process. The EU mode routing through Mistral is practical for firms with data residency requirements. Docker Compose deployment makes self-hosting straightforward.

Who should use this?

Legal tech developers building contract review pipelines should examine the architecture for ideas—even if they don't deploy Lavern directly. Law firms exploring AI-assisted document review might find the debate protocol useful for surfacing competing interpretations. Compliance teams needing EU-hosted inference will appreciate the Mistral routing. At 37 stars with no public benchmark, this is not production-ready for client-facing work. The README is honest: treat the engineering as the contribution and the quality claims as a hypothesis.

Verdict

Lavern is a thought-provoking prototype that demonstrates how multi-agent debate, evidence citation, and layered verification could work in legal AI. The architecture is real, the code is open, and 1,677 tests pass. But with a 0.85% credibility score, minimal community validation, and no independent quality benchmark, treat this as an inspiration source rather than a deployable product. Fork it, learn from it, but don't ship it to clients yet.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.